Why two good laboratories measuring the same tissue get different numbers

A material parameter looks like a fact. It arrives in a report as a number with a unit, it goes into a design file or a finite element model, and from that point onward it behaves like something that was measured rather than something that was produced. For engineered materials that is close enough to true. For biological tissue it is not, and the gap is larger than most people outside a test lab would guess.

What "produced" means

Between a piece of tissue and a number on a page sit perhaps twenty decisions. How the tissue was stored, and for how long. Whether it was frozen, and how it was thawed. How it was kept hydrated during preparation and during the test. How the specimen was cut, and with what. How its thickness was measured - and thickness enters the stress calculation directly, so an error there propagates straight through. How it was gripped, and whether it slipped. How many preconditioning cycles came before the recorded data. What strain rate was used. Whether strain was taken from the crosshead or measured optically on the specimen itself. And finally which constitutive model was fitted, with which optimiser, over which part of the curve.

Each of those decisions is defensible. Each is made differently in different laboratories, usually for good reasons. And almost none of them travel with the number.

The experiment nobody had run

C4Bio, the Community Challenge towards Consensus on Characterization of Biological Tissue, was set up to find out how much that matters. It was initiated by FIBER LABS, the Virtual Physiological Human Institute, and the Avicenna Allience.

The design is deliberately plain. Take one well-defined use case, uniaxial tensile testing of porcine aorta. Prepare samples from a common source. Send them to expert laboratories around the world. Ask each to report the mechanical behaviour of the tissue.

In the first round, 24 laboratories were invited to apply whatever method they normally used. The results showed substantial methodological diversity, and substantial variability in the resulting stress–stretch behaviour. That much might have been predicted.

What happened when everybody agreed

The second round is the interesting one. The participating laboratories developed a consensus protocol together, and 19 adopted it. Sample preparation was standardised. A common cutting tool was distributed so that specimen geometry would be consistent. Thickness measurement was harmonised. The test protocol itself was fixed.

Significant variability persisted.

And it could not be attributed solely to biological differences between the samples. Something about how the measurement is performed (beyond every step that had just been standardised) is still moving the answer.

That is an uncomfortable result to publish, and it is the most useful thing in the paper. It says that harmonising the obvious variables is necessary but not sufficient, and it points the community at the harder question of which of the remaining ones actually matter.

Why this should concern an engineer

If you buy tissue data, or generate it in-house, the spread between competent laboratories is a real uncertainty in your design input - and it is normally invisible, because the number arrives without its provenance.

If you build computational models it is more pointed. Model credibility frameworks (ASME V&V 40, and recent regulatory guidance on computational modelling in device submissions) ask you to justify the evidence your model rests on. The uncertainty in the validation data is part of that case. "We took the value from the literature" is a weak position if the literature values for that tissue span a wide range and none of them carry their protocol.

And if you benchmark a device against published data, you may be comparing yourself against a protocol as much as against a material.

Five questions worth asking your test laboratory

How was the specimen prepared and stored, and is the protocol written down? How was thickness measured, given that it enters the stress calculation directly? How many preconditioning cycles were applied, and does the reported data come from before or after them? Was strain measured on the specimen or taken from the machine? And which constitutive model was fitted, over what part of the curve, and how good was the fit?

None of these is an accusation. Any laboratory worth using will answer them quickly and will be pleased to be asked. A laboratory that cannot is telling you something.

Round three

The third round is now open, and it introduces something new: synthetic samples developed specifically to mimic the mechanical behaviour of arterial tissue. The reasoning is simple. If biological variability is removed from the specimen, then whatever spread remains between laboratories has to come from the measurement itself. It is the sharpest lens the challenge has had.

One last thing

FIBER LABS coordinates C4Bio. We are also a testing laboratory, which means every one of these findings applies to us as much as to anyone who took part. That is uncomfortable, and it is precisely why the work is worth doing. The alternative is a field in which every laboratory quietly assumes its own numbers are the right ones.

Read the full publication here: https://pubmed.ncbi.nlm.nih.gov/41138606/